Add logging through W&B

README.md

@@ -45,6 +45,8 @@ Implementation of _Auxiliary Classifier Generative Adversarial Network_.
 
 Paper: https://arxiv.org/abs/1610.09585
 
+[Trained Model](https://app.wandb.ai/borisd13/Keras-GAN_AC-GAN/runs/tz46c8ow?workspace=user-borisd13)
+
 #### Example
 ```
 $ cd acgan/
@@ -79,6 +81,8 @@ Implementation of _Bidirectional Generative Adversarial Network_.
 
 Paper: https://arxiv.org/abs/1605.09782
 
+[Trained Model](https://app.wandb.ai/borisd13/Keras-BiGAN/runs/2b88b5vv?workspace=user-borisd13)
+
 #### Example
 ```
 $ cd bigan/
@@ -92,6 +96,8 @@ Implementation of _Boundary-Seeking Generative Adversarial Networks_.
 
 Paper: https://arxiv.org/abs/1702.08431
 
+[Trained Model](https://app.wandb.ai/borisd13/Keras-BGAN/runs/360cv3g4?workspace=user-borisd13)
+
 #### Example
 ```
 $ cd bgan/

acgan/acgan.py

@@ -1,5 +1,6 @@
 from __future__ import print_function, division
 
+import wandb
 from keras.datasets import mnist
 from keras.layers import Input, Dense, Reshape, Flatten, Dropout, multiply
 from keras.layers import BatchNormalization, Activation, Embedding, ZeroPadding2D
@@ -21,6 +22,16 @@ def __init__(self):
  self.img_shape = (self.img_rows, self.img_cols, self.channels)
  self.num_classes = 10
  self.latent_dim = 100
+
+ # Log project run
+ wandb.init(anonymous='allow',
+ project="Keras-GAN_AC-GAN",
+ config={"img_rows": self.img_rows,
+ "img_cols": self.img_cols,
+ "channels": self.channels,
+ "img_shape": self.img_shape,
+ "num_classes": self.num_classes,
+ "latent_dim": self.latent_dim})
 
  optimizer = Adam(0.0002, 0.5)
  losses = ['binary_crossentropy', 'sparse_categorical_crossentropy']
@@ -117,6 +128,9 @@ def build_discriminator(self):
  return Model(img, [validity, label])
 
  def train(self, epochs, batch_size=128, sample_interval=50):
+
+ # add extra parameters to log
+ wandb.config.update({"epochs": epochs, "batch_size": batch_size})
 
  # Load the dataset
  (X_train, y_train), (_, _) = mnist.load_data()
@@ -167,6 +181,9 @@ def train(self, epochs, batch_size=128, sample_interval=50):
 
  # Plot the progress
  print ("%d [D loss: %f, acc.: %.2f%%, op_acc: %.2f%%] [G loss: %f]" % (epoch, d_loss[0], 100*d_loss[3], 100*d_loss[4], g_loss[0]))
+
+ # Log progress
+ wandb.log({'D loss': d_loss[0], 'acc': d_loss[3], 'op_acc': d_loss[4], 'G loss': g_loss[0]}, step=epoch)
 
  # If at save interval => save generated image samples
  if epoch % sample_interval == 0:
@@ -188,6 +205,7 @@ def sample_images(self, epoch):
  axs[i,j].imshow(gen_imgs[cnt,:,:,0], cmap='gray')
  axs[i,j].axis('off')
  cnt += 1
+ wandb.log({"images": fig}, step=epoch)
  fig.savefig("images/%d.png" % epoch)
  plt.close()
 

bgan/bgan.py

@@ -1,5 +1,6 @@
 from __future__ import print_function, division
 
+import wandb
 from keras.datasets import mnist
 from keras.layers import Input, Dense, Reshape, Flatten, Dropout
 from keras.layers import BatchNormalization, Activation, ZeroPadding2D
@@ -24,6 +25,15 @@ def __init__(self):
  self.img_shape = (self.img_rows, self.img_cols, self.channels)
  self.latent_dim = 100
 
+ # Log project run
+ wandb.init(anonymous='allow',
+ project="Keras-BGAN",
+ config={"img_rows": self.img_rows,
+ "img_cols": self.img_cols,
+ "channels": self.channels,
+ "img_shape": self.img_shape,
+ "latent_dim": self.latent_dim})
+
  optimizer = Adam(0.0002, 0.5)
 
  # Build and compile the discriminator
@@ -139,6 +149,9 @@ def train(self, epochs, batch_size=128, sample_interval=50):
 
  # Plot the progress
  print ("%d [D loss: %f, acc.: %.2f%%] [G loss: %f]" % (epoch, d_loss[0], 100*d_loss[1], g_loss))
+
+ # Log progress
+ wandb.log({'D loss': d_loss[0], 'acc': d_loss[1], 'G loss': g_loss}, step=epoch)
 
  # If at save interval => save generated image samples
  if epoch % sample_interval == 0:
@@ -158,6 +171,7 @@ def sample_images(self, epoch):
  axs[i,j].imshow(gen_imgs[cnt, :,:,0], cmap='gray')
  axs[i,j].axis('off')
  cnt += 1
+ wandb.log({"images": fig}, step=epoch)
  fig.savefig("images/mnist_%d.png" % epoch)
  plt.close()
 

bigan/bigan.py

@@ -1,5 +1,6 @@
 from __future__ import print_function, division
 
+import wandb
 from keras.datasets import mnist
 from keras.layers import Input, Dense, Reshape, Flatten, Dropout, multiply, GaussianNoise
 from keras.layers import BatchNormalization, Activation, Embedding, ZeroPadding2D
@@ -23,6 +24,15 @@ def __init__(self):
  self.channels = 1
  self.img_shape = (self.img_rows, self.img_cols, self.channels)
  self.latent_dim = 100
+
+ # Log project run
+ wandb.init(anonymous='allow',
+ project="Keras-BiGAN",
+ config={"img_rows": self.img_rows,
+ "img_cols": self.img_cols,
+ "channels": self.channels,
+ "img_shape": self.img_shape,
+ "latent_dim": self.latent_dim})
 
  optimizer = Adam(0.0002, 0.5)
 
@@ -119,6 +129,9 @@ def build_discriminator(self):
 
  def train(self, epochs, batch_size=128, sample_interval=50):
 
+ # add extra parameters to log
+ wandb.config.update({"epochs": epochs, "batch_size": batch_size})
+
  # Load the dataset
  (X_train, _), (_, _) = mnist.load_data()
 
@@ -160,6 +173,9 @@ def train(self, epochs, batch_size=128, sample_interval=50):
 
  # Plot the progress
  print ("%d [D loss: %f, acc: %.2f%%] [G loss: %f]" % (epoch, d_loss[0], 100*d_loss[1], g_loss[0]))
+
+ # Log progress
+ wandb.log({'D loss': d_loss[0], 'acc': d_loss[1], 'G loss': g_loss[0]}, step=epoch)
 
  # If at save interval => save generated image samples
  if epoch % sample_interval == 0:
@@ -179,6 +195,7 @@ def sample_interval(self, epoch):
  axs[i,j].imshow(gen_imgs[cnt, :,:,0], cmap='gray')
  axs[i,j].axis('off')
  cnt += 1
+ wandb.log({"images": fig}, step=epoch)
  fig.savefig("images/mnist_%d.png" % epoch)
  plt.close()
 

requirements.txt

@@ -4,6 +4,7 @@ matplotlib
 numpy
 scipy
 pillow
+wandb
 #urllib
 #skimage
 scikit-image